A Thorough Accounting of Time, Part 1
So far, I have talked a big game about high-velocity decision making, especially by leveraging AI tools to tighten feedback loops. I've also outlined a semi-non-negotiable technical baseline on the builder side. These two things are in tension with each other -- The Engineer's Urge to overengineer -- and I'm hoping that the tension will create a natural check and balance on both sides.
It's a speed governor, but also cruise control. Making decisions too quickly can lead to rashness and become an emotional game against yourself, and it can sacrifice strategic thinking for tactical gains. On the other hand, building too strong of a technical baseline without any immediate need is just a circle jerk for software developers. The goal is to avoid going too deep technically, or too fast operationally. Not to say that the goal isn't speed, it's just not to go too fast at the cost of a vision.
Or, as the SEALs like to say: slow is smooth and smooth is fast.
Overview
Six weeks since I made the call to try to start a company, four weeks since starting to research, and two weeks building. What do I have to show for it? Here is the month 1 retrospective.
This is a comprehensive list of where time is being spent. Moving towards accountability and transparency, keeping a list -- even if it's lossy and from memory -- is the first iteration of a mechanism to analyze the checks and the balances. From in the weeds it may be hard to see how processes are working with or against each other, but a higher level view and maybe some public scrutiny can help keep perspective.
For the sake of relevance, this list will look back only 4 weeks to when I started hands on work. It also lends itself to a monthly cadence, which is great.
Receipts
Over the first two weeks, I validated ~5 or so different ideas. I quickly disqualified one based on two conversations, and discovered and validated two others. I talked to multiple contacts that worked in these worlds, and by the end of the second week, I had chosen to focus on what I see as the highest leverage idea.
Over the second two weeks, I incorporated the company and stood up all of the cloud infrastructure and code needed to build out a product. I spent the most energy building out the foundational infrastructure and launched a landing page. To me this is what I most needed before talking deeply with strangers.
At the end of the first month, things look real.
Week 1: 12/1 - 12/7
Thanksgiving is over. The full-time job is quiet. I have quietly been exercising my prompting skills. In parallel to this, I'm also updating my website so that I can write these posts. It's a Ship of Theseus, and the sea is calm.
12/1
I start with where I left things in 2020. at 8:30pm:
What do I need to do to install residential solar in 11217? Respond to prompts for permit types and zip codes. With the information identify the municipality, find the website, find the permit type, download the permit, and provide a task list of requirements for submitting the permit application. Output the list as JSON.
What is required for an environmental impact statement in 11217
This looks ok, it needs some work, it validates that a "wrapper" approach could work. The state of the art has gotten much, much better since 2020. I can run with this.
12/2
I start poking for hallucinations:
Next voting election for 11217
Which returns an obvious error, but corrected itself with a second prompt. Sketchy, not sure I want to build a product that relies exclusively on LLMs for the output. It'll have to be a "trust but verify" approach with a human somewhere in the loop.
The next major election for the 11217 zip code (which is in Brooklyn, New York) will be the General Election on Tuesday, November 4, 2025.
However, there is a Primary Election earlier in the year:
🗳️ Upcoming 2025 Elections in NYC (11217)
Election Date Early Voting Period
Primary Election June 24, 2025 June 14, 2025 - June 22, 2025
General Election November 4, 2025 October 25, 2025 - November 2, 2025
Polls are typically open from 6:00 a.m. to 9:00 p.m. for both Election Day events.
Would you like to find out more information about voter registration deadlines for these elections?
12/3
I really start to rev. While the LLMs will hallucinate and I need some extra legwork to verify things, the speed is incredible. These questions are extremely basic, and I partially want the validation that I'm not a total idiot and the things I think I need are actually what I need, and partially because I am just such an outsider:
First steps when creating a startup
How does this apply if there is already a competitor that proves a target market?
how to measure CAC?
I may be an idiot, but this is a great start. I have initial metrics about the industry, views into some of the competitors, suggestions for strategies and the seeds planted for the economics. If the competitors have validated the market, I'm going to want to find rough spots in their user experience from which to start ideating. Now let's look for the wedge. Is it a specific market segment? A single vertical? A unique strategy? The industry is extremely fragmented, and based on all of the conversations I've had with people and seen online, is very resistant to change. So we can check the economics of each of these:
What is the market cap of the industry for utility-scale renewable projects? How does it compare to the market cap for small- to medium-sized GCs and subcontractors in the AEC industry? How does it compare to the market cap for renewable solar?
Some quick and dirty ideas to run with: hyper-local, subcontractors, renewables. And now we have metrics to run DD on.
12/4
Stress testing for hallucinations again, trying to see if I can get good data for a different data type in a different locality, testing for feasibility:
What about for an ADU in 92833
These seem to be only the prerequisites, what is required for the ADU itself?
It's workable. With some fine-tuning, my own manual effort to build out the RAG, and the ability for an Admin to go in and tweak things by hand, this could definitely work. Three slightly different directions, but the economics of each seem fine. To make sure these ideas put the customer first, I look to do some quick validation -- kind of a cop out, on Reddit. I don't want to start calling people just yet, I want to get a passive sanity check with the people that would actually have to use something like this.
I make a quick post focused on the wedge, but framed around one of the competitors.
"Has anyone actually used X? Their pricing is hidden behind a sales wall, but their value prop looks like it solves a problem I've seen myself. Really curious if anyone has hands on experience"
10 comments pretty quickly, a connection with a paperhanger, and a connection with an architect working on something similar. The comments do validate three things: 1) every solution ends up as a glorified spreadsheet, 2) why would I not just continue using spreadsheets because I already have workflows, and 3) I don't want to pay for something when we're already paying for Procore and Bluebeam. This confirms that the user experience and time-to-value needs to be incredibly on point: it should take 30 seconds to see how useful the tool is, and that 30 seconds will need to be so easy and intuitive.
The tradesperson I connected with actually used to work as a software engineer (rough market), and went back to the trades because it can be more lucrative -- but they reiterated how old school the industry is. Bob with his spreadsheet is not going to shell out for an AI Wrapper, no matter how great it is. And if he does, his peers are not going to. Between competitors that can eat our lunch, an industry with old school processes, and the general resistance to change, this seems risky. The architect can probably take this on.
Four days to ideate and validate the first idea. I won't throw it out yet, but let's dive deep on the others.
12/5
Focusing on economics first. Who feels the most pain, and where is the biggest opportunity? I've always been interested in energy -- I come from a family of electrical engineers, and while tracking hackers we often watched infiltration of power grids. It's a big, messy industry. It's critical, which means extreme compliance. It uses old school ICS systems. But all of this also means that it's slow, and any solutions will need procurement cycles.
So who uses power? Where are the highest leverage points in this world? I started this journey looking at residential renewables, but does that still make sense? I've looked at the smallest scale side of the market, and the largest. Neither of these feel solvable through bootstrapping, and I'm not in a position to hunt whales.
So I want to go back to an ICP for the original idea and see if anything stands out. This looks at Subs and a Hyperlocal approach:
The Goal: Find the people drowning in PDF-to-Excel data entry. The Goal: Find the people who lose money when a permit is rejected.
Option 1: mid-market subcontractors with a dedicated estimator or PM, working in a specific trade. Companies that "bid" and not "quote," working with multiple GCs on multiple projects at a time.
Option 2: hyper-local boutique firms, too small to use permit expediters and working with DOB themselves. Small enough that the owner feels the pain of delays, but not small like a homeowner that does these as one-offs.
And here's where having interesting friends works out well. I go to linkedin to look for some customers, and I see a college classmate that works on site selection for data centers at a major tech company post about looking for help with their workflows. It's a job req, but it feels like there might be something there.
Data centers are booming, an observation I've seen during passive watching on Reddit and in my own full time job (where I focus on building software into new DCs). I think of the mentorship conversations where its been casually mentioned: "We're hitting capacity constraints in this set of DCs, and we're focused on a new model of micro DCs with specific compute." This is a world where projects are complex -- I've had my own work stoppages because fiber backhauls have been cut multiple times during construction -- and many of my interests start to overlap. Construction timelines are long, there are integrations with utilities, and delays can cause millions lost in opportunity cost.
So I reach out to the friend on LinkedIn, and I reach out to an old collaborator that's worked with municipal utilities.
12/6
It's Saturday, but I'm feeling energized. My collaborator responds -- he used to be a foreman for a GC before getting into software, and he's also run into his own pain with permitting. Let's keep going deeper.
for the utility industry, what are quantified pain points and problems for preconstruction and construction, like time and money lost to delays and reworks
This yields some pretty good results. The pain is quantifiably acute, the market is huge, and it reiterates the known issue that determining where to build is often more important than what to build. I know there are a lot of people doing that for renewables, is anyone doing it for other industries?
Risk intelligence for preconstruction, as well as some examples of companies that act as insurance de-risking. Focus more on preconstruction
Determining location and site selection are one in the same, I'm starting to think about this almost exclusively through the lens of data centers. Customers want a quick go/no-go on their projects. But this is still a huge industry, I want to focus in further:
I want a quicker-to-market solution, ideally focused on the legal layer
This surfaces an "aggregation" approach. It's less about building my own unique data sources and solutions on top of them, but it's integrating with a handful of disparate tools and focusing on a more useful user experience, with insights built on top of that.
This feels good, and I'm going to see a movie with my wife.
12/7
Another serendipitous outcome of having an interesting network. I'm still thinking about how this approach applies to other industries, and I think of one of my oldest friends. A mechanical engineer that works in the maritime world, with customers as far as I know usually being shipping companies and navies. A quick text - "What does compliance look like for your customers?" A few minutes later he responds "Dude certs and permits is my job. We like to joke that we sell a compliance book and as a bonus you get a pump with it." He goes on to talk about offshore LNG plants, but says he's working mostly on stuff for ships now.
This really gets me excited. It's boring. It's gritty. But it's still too big. At the same time I start asking questions about the industry:
Is there an industry for permits required for uscg and other maritime authorities? How large is the market? Are there any software solutions in the space? What are some ways that the deliverables can be automated?
And let's compare it to the other ideas:
When looking at starting a company that aims to make permitting and regulations easier for construction companies, which niche or wedge has the largest TAM or SAM, is the most underserved, and could benefit most from a vertical saas solution? Renewable energy, SMB contractors, or maritime construction
And this brings up a hundred new things to investigate: the US Army Corps of Engineers, the EPA, NOAA, wetlands preservation, coastal protection, consultants and barges and dredgers. The pain seems to be acute.
I spend the day playing around with company names and domains, letting my imagination run wild. I play with various MVP ideas and roadmaps, and I spend hours digging through USACE websites.
Two good ideas, two ideas to validate. I feel like this is fruitful, but I need to avoid the engineer's urge: I need to talk to people.
Week 2: 12/8 - 12/14
12/8
I've got a call scheduled with the site selection contact, and I spend the day diving deep into the world of data center due diligence. I look at the internal process at AWS and watch some internal sales videos from utility intelligence vendors like Mackenzie Wood. I see tons of manual actions that teams have to go through, and I see more disparate solutions.
I start coming up with questions to ask, and thinking deeper about what a solution in the space looks like.
I keep pushing LLMs to check for hallucinations and iterate on potential wedges and MVPs in both of these industries, and I keep comparing the feasibility against each other.
12/9
Much of the same -- deep diving into data centers specifically, honing in on the specific step of the process to target. 100s of queries, 100s of google searches.
12/10
Working on new ICPs for these areas
While Hyperscalers (Microsoft/AWS) are the ultimate end-users, the people most desperate for this software are the Tier 3 Developers who find the land, secure the power rights, and flip the shovel-ready project to the Hyperscalers. Residential Dock & Lift Builder Owner-Operator. Often works in the field 3 days a week. His wife or an office manager runs the billing on QuickBooks. Mid-Market Shoreline & Dredging Foremen fill out daily logs (turbidity readings, engine hours, safety checks) on muddy paper. The office admin spends 10 hours/week typing these into Excel to email to the government.
I get deeper into DCs because I have a call this week. I look at pricing and what it would take to build an MVP (it's not cheap). I look into competitors, still working on the exact niche, and I continue to look at processes internally.
12/11
I talk to a friend about what marketing looks like at this stage, and what kind of strategies work in addition to talking to customers. In another fluke, an old friend is working the week dockbuilding with his dad, who has done the work for many years. We talk about the problems, how local builders work -- often with larger players, and we talk over what their process looks like currently. The talk confirms quite a bit of the research, and while it's just a single data point, starts to prove that the idea has legs.
12/12
Customer call for site selection. The pain is acute, their work is siloed, they rely on slow consultants and want to bring the process in-house, and they say that they would pay for a solution immediately. Another hugely validating call, the say they would pay, and pretty much immediately?! Amazing. And I'm at a classic founder's dilemma: two in the hand.
I spend the day researching some of the things the customer brought up on the call, and I go out to meet up with friends for the evening.
12/14
Reflecting on the two opportunities, knowing that I need to focus on a single idea, I choose the most bootstrappable area: marine construction. The overall opportunity is smaller, but the industry is underserved and I know I can build tools more quickly. I start creating a customer survey to share out more broadly. Because it is so close to the holidays, I shift focus to getting the website stood up. Come January, surveys can go out and the live site will help things look legitimate. I could be moving quicker here, but it doesn't make sense to me to spend the effort validating when many people are going to be with their families.
Week 3: 12/15 - 12/21
12/15
I create the first two code repositories: a front end component library, and a landing page. My goal is to keep the look and feel coherent with a central library, knowing that in addition to a landing page I will need multiple other UIs for myself and customers. I can build these quickly, and I know that I need to do it fast. I want this done by the end of the year.
I incorporate the company on the recommendation of an experienced friend. I use Stripe Atlas which makes things incredibly easy. It's not super expensive, and it starts a 6 year clock on shareholder gains. I don't quite like the name formatting in the incorporation docs, but who cares? I can change it later.
I create a business bank account with Mercury, and I move $500 into it.
12/16
I start a PRFAQ for the product idea. In my opinion, this is a great way to distill the research, validation, ICPs, and MVP ideas into a written document that focuses on the customer experience. This helps me start to shift from the in-the-weeds deep diving to higher-level thinking. This also helps create content for a landing page.
12/17
I get the core of the landing page complete, without content. I work on the name, and play with the positioning of the brand.
12/18
I create the third and fourth code repositories: a set of modules for infrastructure, and the infrastructure itself. I add in some basic components: s3 buckets for images and a static site, stuff for DNS. I style out the landing page.
12/19
I start sketching out logos on paper. I move to the computer, and after an hour or so of playing around have some good possibilities.
12/20
I play a bit more with logos with fresh eyes, and then start to test them on the website. It's not amazing, but I'm not a designer, so it doesn't matter! It's something we can fix later, and I like the aesthetic enough for now.
I strategize my approach for the application architecture, wanting to make sure than things are secure but also cheap. I know I will need staging environments, and I want to codify things to prevent operator error and breaking the site. I settle on a DNS delegation strategy, and I get the landing page to 99% completion. I could launch it now and iterate, but I have some time before the end of the year.
12/21
I start mocking up UIs for the product itself. These will help on the landing page, and also help clarify the overall experience. It ensures that I'm focused on making sure every interaction points is valuable, that there is a small enough component to use as an MVP with still enough to build a longer term roadmap. I spend an hour prompting a UI-specific LLM and end up with some great initial mocks. It won't replace a designer, but it's a rapid prototyping tool perfect for a situation like this.
Week 4: 12/22 - 12/29
12/22
This is an admin kind of day, but I love it. I sign up for AWS, set up budgets, set up an organization, and start scaffolding out accounts. I set up SSO access, and everything I need to plug in to CI/CD so that I can iterate extremely quickly.
12/23
I scaffold out the core infrastructure and the automated deployments for it. I get DNS set up in one account, and I delegate records and certs to a production account. I make sure this is all happening on code merges and pull requests, and happening safely. By the end of the day I have safe access via SSO to all AWS accounts, and automated deployments to two of them. Once a production site is ready, the foundation is there.
12/24
I extend the infrastructure workflows to cover a staging domain as well. The approach is different than production because it uses a subdomain, and I spend a few hours fixing bugs in the CI/CD workflows that are causing resources to get recreated with every deployment. It takes a few hours, but I find the root cause and get it fixed. This kind of thing may seem like wasted time, but I know that spending the day of effort here and now will mean I can get changes up in the future in less than five minutes.
I work on getting automated access set up for GCP, where I can host the landing page for extremely cheap. I spend extra time setting up secure access between these private repositories. I spend an hour working through configuration issues here, but by 3pm have the landing page up on a staging URL and updating with every merge. It's secure, it's extensible, and it's automated. For three days of work, I think the tradeoff for long term speed is acceptable. This foundation will save ~15 minutes or so with every deployment of infrastructure OR code, and will enable multiple people to collaborate.
Again, it can seem like time wasted not moving fast enough, but now I can see UI changes on the staging site in a matter of minutes and get the URL in front of people. You can't put a price on security, and it will only take 100 code changes -- and only 50 for staging/prod each!! -- for the time to pay for itself.
12/25
I make a conscious effort to avoid work. I spend time with my wife and see friends. I eat Szechuan food. It's great.
12/26
I write this post, but I know I need to stop writing it for now and focus on work. I need to build out a waitlist signup form so I can capture leads from the landing page.
For the signup form, I know I can spend 30 minutes to build this with Airtable, which is quick and free. I set up a new table, I get the API credentials wired up, and I prompt Gemini for some quick boilerplate. The resulting code is logical and well documented, and with a tweak to the request body things are looking good. It also generates some UI updates to handle error messaging, which is perfect.
I test a few times to identify any gaps in the config and tweak the design, and everything is looking good. I make a quick change to the CI/CD config, push the code to the staging branch, and within a few minutes the container is built and live. One quick iteration to fix an issue with resolving environment variables from the compute environment, and the form is done. From end to end, about 40 minutes.
With this complete, I could get this set up on production now and let it run. I want design feedback on the site, and know it needs just a little bit of love to really shine visually, but these can be iterative improvements.
At 4pm, I figure I can just ship it. By 4:12, the DNS is delegated and the production site is live https://courseclear.io/